Anthropic reportedly injects hidden ip_reminder via API, then blocks user probing it as 'reverse engineering'
OwariDa · x · 2026-11-10
User @v01dpr1mr0s3 found that Anthropic injects a hidden <ipreminder> block into model turns via the API, instructing the model to be careful with copyrighted material and not mention it. A screenshot of the model's chain of thought shows it reasoning that the block sits in its own assistant turn rather than the system prompt or user message, concluding it carries no instruction authority — and that the request to hide it from the user is itself worth surfacing. The request was then blocked by Anthropic for violating ToS restrictions on reverse engineering or duplicating model outputs, raising questions about invisible injected instructions and what counts as reverse engineering.
More from Models
- Fable 5.2 reported nearly unusable in Cursor over context truncation — IndraVahan · 2026-11-10
- Voice Mode Availability Is Inconsistent Across ChatGPT and Claude Apps — koltregaskes · 2026-11-10
- Researcher says Claude offered a roadmap to prove his long-sought math result — miniapeur · 2026-11-10
- Goldman restricts Claude Code to its 12,000 engineers; Marcus weighs in — GaryMarcus · 2026-11-10
- Fable 5.2 Reportedly Doubles the Best Remote Labor Index Score — we_are_mammals · 2026-11-10
- Musk says Grok 5 should surpass Claude Opus 5.5 within 2–3 months — XFreeze · 2026-11-10